This project responds to timely methodology needs and opportunities in biomarker evaluations for cancer risk prediction, diagnosis, early detection, and prognosis In the Early Detection Research Network (EDRN). Nevertheless, our work will have implications more broadly for studies of medical tests. The first aim proposes the development of ROC analysis that allows combination of datasets from multiple studies, populations or platforms including determining if ROC performance is common across datasets; providing efficient estimation of the common ROC performance; and quantifying factors either within or across datasets that affect ROC performance when performance is not common. The proposed methodology provides a unified framework for risk prediction modeling and ROC regression methdologies that have been disjoint and potentially inconsistent. A second aim is concerned with group sequential design methods development with the goal of preserving high quality biospecimens in biomarker validation studies, including developing methods for estimating the performance of single markers or marker combinations that accommodate the potential for early study termination; and investigating optimization of design parameters such as timing of interim analyses and criteria for early termination. A final aim responds to methodology needs in study designs for prognostic biomarker evaluations, including efficient ways of selecting for ascertainment of data and specimens; estimating a variety of biomarker performance metrics from these designs; and comparing them with existing designs and estimation approaches. The general approach involves the specification of pertinent statistical methods and the use of both theoretical probablitistic methods, and computer simulations that are informed by application to data from substantive research contexts in which the four participating biostatistical investigators are engaged.